2011-01-14 05:39 wpwrak: http://sources.redhat.com/bugzilla/show_bug.cgi?id=3266 2011-01-14 05:39 so much for your love of GNU 2011-01-14 05:52 lekernel: good old uli ... one of the reasons why glibc has been forked 2011-01-14 05:55 lekernel: well, "forked". they still track glibc, so technically, it's not a fork ;-) http://www.eglibc.org/faq 2011-01-14 06:07 he dash vs bash issues :-) 2011-01-14 06:08 that's not the point, the point is that the libc doesn't have to dictate a particular shell 2011-01-14 06:09 another example of GNU imperialism, and then they blame Microsoft for the same thing 2011-01-14 06:09 lekernel: are you using glibc or eglibc ? in the former case, i'd strongly suggest to consider switching. 2011-01-14 06:10 lekernel: there was also a connection with the binutils. in some cases, you can get "bad" binutils if you use the official (Uli) versions. 2011-01-14 06:10 is there any non-GNU replacement for those binutils btw? 2011-01-14 06:10 for gcc there's llvm, which is good 2011-01-14 06:10 but it still uses the gnu assembler and linker :( 2011-01-14 06:11 GNU is whole OS, remenber that ;-) 2011-01-14 06:12 well, you can call it imperialism 2011-01-14 06:14 lekernel: I don't know of any alternative to binutils 2011-01-14 06:14 mh... me neither... again something that would need to be addressed 2011-01-14 06:15 the binutils code quality isn't much better than gcc's 2011-01-14 06:15 lekernel: gnu coding style per se sucks :) 2011-01-14 06:19 nah, and i was reading their codying style guide :/ 2011-01-14 06:21 I didn't know they had a coding style guide 2011-01-14 06:22 found it, and the first thing I read is: 2011-01-14 06:22 Please dont use win as an abbreviation for Microsoft Windows in GNU software or documentation. In hacker terminology, calling something a win is a form of praise. If you wish to praise Microsoft Windows when speaking on your own, by all means do so, but not in GNU software 2011-01-14 06:22 stupid... 2011-01-14 06:25 lekernel: the first thing you read is the trademarks section ? ;-) 2011-01-14 06:25 well, they put it at the beginning 2011-01-14 06:25 1 About the GNU Coding Standards 2011-01-14 06:25 2 Keeping Free Software Free 2011-01-14 06:25 2.1 Referring to Proprietary Programs 2011-01-14 06:25 2.2 Accepting Contributions 2011-01-14 06:25 2.3 Trademarks 2011-01-14 06:26 it's just bait ;-) 2011-01-14 06:26 i'm sure they enforce it 2011-01-14 07:00 lekernel: when was the last time you saw a troll who didn't enjoy getting challenged ? ;-) 2011-01-14 08:32 [commit] kyak: Tile: the 15 Puzzle game written in Qt4 http://qi-hw.com/p/openwrt-packages/bd53759 2011-01-14 09:47 lekernel: maybe you could port the BSD linker? 2011-01-14 10:44 DocScrutinizer: can i borrow your vast experience with troublesome electronics for a moment ? 2011-01-14 11:26 [commit] kyak: Tile: check if the puzzle is solved http://qi-hw.com/p/openwrt-packages/ca4ebe4 2011-01-14 11:33 wpwrak: sure 2011-01-14 11:34 DocScrutinizer: kewl. given a little ~2.4 GHz transmitter, set to emit an unmodulated 2.45 GHz test signal 2011-01-14 11:34 just let me get a coffee 2011-01-14 11:35 DocScrutinizer: the signal goes via the air to a spectrum analyzer (well, my USRP2, so it's uncalibrated) 2011-01-14 11:35 DocScrutinizer: this it what it looks like: http://downloads.qi-hardware.com/people/werner/wpan/tst-cw/ 2011-01-14 11:35 DocScrutinizer: the interesting examples would be in the 4th row (the one with two graphs) 2011-01-14 11:36 DocScrutinizer: ah, the signal is offset by 0.5 MHz. so it's 2449.5 MHz in the left picture and 2450.5 MHz in the right picture 2011-01-14 11:37 DocScrutinizer: you can see that I have about 40 dB between the peak and the trouble underneath. does this look halfway reasonable so far ? 2011-01-14 11:46 wpwrak: the assumption "unmodulated" seems incorrect. Esp row 5 you seem to have *massive* modulation by pink noise 2011-01-14 11:46 DocScrutinizer: wait .... row 4 first :) 2011-01-14 11:46 (well, maybe actual data even) 2011-01-14 11:47 DocScrutinizer: does row 4 look acceptable for an unmodulated carrier ? 2011-01-14 11:47 same, but to a lesser extent 2011-01-14 11:47 not really 2011-01-14 11:47 could you compare to other TX hw? 2011-01-14 11:48 to rule out your RX has some intermodulatio with noise on e.g. power supply 2011-01-14 11:48 DocScrutinizer: only to other of my designs using the same chips. they are in rows 2 and 3. they all look quite similar. 2011-01-14 11:48 DocScrutinizer: ah .. i see what you mean. hmm. 2011-01-14 11:49 I see. What about some stupid babyphone or whatever 2011-01-14 11:49 DocScrutinizer: don't have that. besides, it would be modulated. 2011-01-14 11:49 and check your power suplies 2011-01-14 11:50 a simple babyphone wouldn't be modulated when you're hush 2011-01-14 11:50 :-) 2011-01-14 11:51 DocScrutinizer: oh, and my USRP2 has a resolution of only 14 bits. that would be ~42 dB. 2011-01-14 11:51 or lemme put it that way: it's modulated 0% 2011-01-14 11:51 oh, so maybe it's quantisations noise? 2011-01-14 11:52 would make sense. You need that babyphone :-) 2011-01-14 11:52 DocScrutinizer: the noise floor I see without any signal is about -70 dB. but i don't know what becomes of a signal. 2011-01-14 11:52 row1 is nice but useless 2011-01-14 11:53 exactly 2011-01-14 11:54 there's just one way: try to 'calibrate' wit a 'known good' signal 2011-01-14 11:55 alas, i don't have that :-( my transceiver board is only half-duplex. 2011-01-14 11:55 of course, i could try to bounce the signal off the moon. that should give me enough time to switch from tx to rx ;-) 2011-01-14 11:55 use an arbitrary wlan or BT or whatever is sending in roughly the same band 2011-01-14 11:55 wlan is waaaay to wide 2011-01-14 11:56 take in account possible modulations. You still can compare what you see to what you expect 2011-01-14 11:56 at this resolution, i just see the whole display jump up when wlan does something 2011-01-14 11:56 :nod: 2011-01-14 11:58 the transceiver is specified with about 48 dB between center and 2nd and 3rd harmonics. 2011-01-14 11:58 so i think i'm at least in the ballpark. 2011-01-14 11:59 mhm 2011-01-14 12:00 yes, that would match, though of course that'S side band, not harmonics 2011-01-14 12:00 yup. they don't specify sidebands 2011-01-14 12:01 NB I don't think row 4 has any particular problems 2011-01-14 12:02 great. that's what i was hoping for :) 2011-01-14 12:02 it just shows to some extent the same noise as row 5 2011-01-14 12:02 at least compared to what follows ... 2011-01-14 12:02 the same pattern of noise 2011-01-14 12:03 ben#3 and ben#4 are unbearable 2011-01-14 12:03 now, row 5. that's the guru punk version of row 4. high on top of a mountain and with spiky hair. 2011-01-14 12:04 I suggest you check your VDD 2011-01-14 12:04 i tried to tame it down a little. first http://downloads.qi-hardware.com/people/werner/wpan/rework-20110112/ 2011-01-14 12:04 then http://downloads.qi-hardware.com/people/werner/wpan/rework-20110113/ 2011-01-14 12:04 i beaded VDD and also doubled the bypassing, with no effect at all 2011-01-14 12:05 what did improve things a little was changing the clock voltage divider circuit 2011-01-14 12:07 note that the rework-20110113 series is tethered. some of my rework made the board too thick to go into a ben. 2011-01-14 12:07 i'm now reworking it to make it thin again, so that i can see if the improvements remain when connecting to a regular ben. 2011-01-14 12:08 some of the measurements suggest that a "full" ben may add some gremlins to the mix, compared to the bare pcb. may be the lcm. 2011-01-14 12:10 yep 2011-01-14 12:11 you are fully aware about the principles I guess. Meassure the delta-f between center and sideband peak and you got the AM signal's frequency 2011-01-14 12:12 for FM it's much harder to calculate and I don't really understand how to do that. I also never understood how to use those FM sythesis keyboards (DX-7 ?) 2011-01-14 12:13 i'm more concerned about the "mountain" for now. i've made the sidebands go away once by adding a small bypass to the rf feed line. 2011-01-14 12:15 so i have about +/- 1 MHz of fairly persistent noise. 2011-01-14 12:16 What I'd do is: use a usual needle (or 1"/25mm wire), extend it with a non-conductive handle of at least 30cm so you can hold it without interfering with the electrical properties. Then touch all pins of your circuit, and whatch noise floor while you do that 2011-01-14 12:18 does the needle connect to anything ? 2011-01-14 12:18 nope 2011-01-14 12:18 just a small "antenna" to catch the noise of your environment 2011-01-14 12:21 I guess you'll find some pins that are massively susceptible to this kind of manipulation (OSC-IN ?). It's worth to have a closer look to those details in your design, maybe use a better trace layout etc for those pins. 2011-01-14 12:22 Also I read to tried to do something to VDD by adding a 1uF. Probably worthless, better add a 1nF, maybe even several. Try different locations to add. Try a bead in VDD 2011-01-14 12:24 one thing that might give you dispair though is the fact that cheap cheesy TX circuits are susceptible to cating up noise *by the antenna* 2011-01-14 12:25 i tried a VDD bead already. didn't do anything. 2011-01-14 12:25 there's not much you could possibly do in that case, except reducing ambient noise (shield converters, e.g. of LCD backlight, etc) 2011-01-14 12:26 if you want you may toss over the URLs to schematics pdf, and layout pdf 2011-01-14 12:26 the antenna is grounded for now. (when i go tethered, i cut the antenna a bit and solder an U.FL connector that grounds the antenna) 2011-01-14 12:27 schematics are here: http://projects.qi-hardware.com/schhist/atusd/ 2011-01-14 12:28 now that i changed the clock divider. it's now 5.6 k : 1 k resistive, followed by 1 nF DC block. 2011-01-14 12:28 i tested the parameter sensitivity of the DC block - 10 pF and 100 nF yield basically identical results. 2011-01-14 12:28 aaah, I seem like recalling there were problems back in good ole' '76 with CB (27MHz), the antenna when detuned could cause the TX power drivers of the walkietalkie to emits all kinds of nasty sideband and harmonics 2011-01-14 12:29 also, dropping the resistive pair by one order of magniture (560 R:100 R) doesn't change the result. 2011-01-14 12:30 the 201101013 series eliminates the antenna from the equation :) it all goes straight to the USRP2. well, through a 10 dB attenuator. don't want to fry the LNA :) 2011-01-14 12:31 (although it's probably fine with the meager 3 dBm i can throw at it) 2011-01-14 12:33 werner, are you *sure* the AT86RF231 doesn't have an always-modulated RF signal? like filling in soem padding when no data is transmitted 2011-01-14 12:33 it's in "continuous wave" test mode 2011-01-14 12:34 besides, the "row 3" result, with 40 dB separation, is with the same chip 2011-01-14 12:34 CW doesn't necessarily mean no modulation 2011-01-14 12:34 aah, ok 2011-01-14 12:36 now .. let's try joerg's bad finger ... 2011-01-14 12:37 hehe 2011-01-14 12:37 touching XTAL1 directly crashes the chip. well, i had that before with the scope probe. 2011-01-14 12:37 don't stab anything :-P 2011-01-14 12:37 YAY 2011-01-14 12:40 touching before the blocking cap raises the noise beyond +/- 1 MHz by about 5-10 dB but doesn't affect the central hill. 2011-01-14 12:41 expected 2011-01-14 12:42 the blocking cap would obviously effectively filter out oise <1MHz 2011-01-14 12:42 noise* 2011-01-14 12:43 maybe .. don't know the input impedance of XTAL1 2011-01-14 12:43 but the more it filters, the better ;-) 2011-01-14 12:44 touching VDD doesn't seem to have much of an effect. btw, neither does my body hovering some 10-20 cm over the circuit 2011-01-14 12:46 aaah, always hard to follow what you do, unless given component names. The whole thing is externally clocked? Yuck! Check that clock for jitter and noise. Consider using a local xtal instead, even if only for ruling out it's the ext clck that is polluted 2011-01-14 12:46 btw, the sidebands vary by around 5 dB while the "hill" remains constant most of the time. sometimes everything jumps, but that may be some wlan again (i've shut down mine) 2011-01-14 12:47 yes, local xtal will be the next thing to try. the USB boards (row 1-3) all have a crystal. 2011-01-14 12:47 of course, a crystal is something i'd really love not to have on the bom :) 2011-01-14 12:48 besides, it would make the board larger. yet another reason to dislike it. 2011-01-14 12:48 using external clocks for TX is usually a bad idea, unless you have a really high quality known good master clock generator somewhere in your circuit 2011-01-14 12:49 the ben's clock should be accurate enough. also, at accutracy of at least the long-term frequency (measured over minutes) is more than sufficient 2011-01-14 12:49 if you must use this external clock, then FILTER it, use R-C bandpass of 3rd order, use L-C tuned to the freq of clock 2011-01-14 12:50 it's all about jitter and noise 2011-01-14 12:50 i tried to use my function generate to provide an external clock, but that didn't go well. see the two "combs" around the middle of http://downloads.qi-hardware.com/people/werner/wpan/rework-20110113/ 2011-01-14 12:51 :nod: 2011-01-14 12:51 the long cables and external gear introduced lots of noise 2011-01-14 12:52 yup. if we discount the sidebands really close to the center, then the "hill" dropped by 5-10 dB, though. but i'm not sure if this is a meaningful interpretation. 2011-01-14 12:54 It's quite obvious now: you need a better clock 2011-01-14 12:54 i tried to find any disagreement between the external clock input and the transceiver's clock output (it can output the divided input clock), but that wasn't very conclusive because i couldn't get a good trigger 2011-01-14 12:54 it's the two blurry curves on http://downloads.qi-hardware.com/people/werner/wpan/rework-20110112/ 2011-01-14 12:56 now, what would be the best way to go about this ? 2011-01-14 12:56 (poking around some more with your "bad finger". nothing really responds to it) 2011-01-14 12:59 get an xtal :-D 2011-01-14 12:59 grmbl 2011-01-14 12:59 I know I know 2011-01-14 12:59 xtal doesn't convey the cool simplicity i'm after 2011-01-14 13:01 any simple LC filter that could help ? right now, i have 1 C and 2 R in the clock path. i should be able to add 2 more components. 2011-01-14 13:01 maybe 3 if i really squeeze things. 2011-01-14 13:02 of course, if the filter does the DC blocking too, that's one more "free" component 2011-01-14 13:07 L-C with high Q, loosely coupled to clock source, might do. Have fun tuning it ;-D 2011-01-14 13:07 That's beyond my competences 2011-01-14 13:07 grr :) 2011-01-14 13:08 that would be a bandpass then ? 2011-01-14 13:12 that would be an L-C parallel resonator tuned to the 16MHz it shall provide, with high Q, means it shall keep oscillating for >> 1/1Mhz, i.e. 50% of sine amplitude @ 16MHz after nn microseconds. Maybe undampened by feedback of XTAL2 to the LC with attenuation so it just doesn't start oscillating itself 2011-01-14 13:13 then gently feed energy from external clock, so an oscillation builds up after several nn microseconds 2011-01-14 13:14 muuch more icky to design, to tune, to debug. And BOM won't get any nicer 2011-01-14 13:14 compared to a cheap XTAL 2011-01-14 13:17 there are no xtals that are small and cheap :-( 2011-01-14 13:17 umm 2011-01-14 13:17 pries open his BT usb stick 2011-01-14 13:18 even the 3.2 x 2.5 xtal i'm using is still around usd 0.6 @ 100 2011-01-14 13:18 yep 2011-01-14 13:20 new data point at the bottom of http://downloads.qi-hardware.com/people/werner/wpan/rework-20110113/ 2011-01-14 13:20 this is back over the air, with an external antenna 2011-01-14 13:21 hmm, this BT stick has a 5 * 3.1mm "16.000 H.ELE.7L" 2011-01-14 13:21 the sidebands almost at the center are probably just antenna tuning. in fact, as it watch it, they're merging into the center 2011-01-14 13:22 that's not 7L that's XXXL ! ;-) 2011-01-14 13:23 well, at least that one is nicely low-profiled 2011-01-14 13:23 even the nearby chip is thicker 2011-01-14 13:23 updated 20110113 again, with the merged sidebands 2011-01-14 13:23 about 25 dB between peak and noise floor. getting close :) 2011-01-14 13:24 my goal is to get something around 30 dB (without crystal) 2011-01-14 13:25 802.15.4 has some requirements on the channel width, but i think i comply with them even with the original (horrible) circuit 2011-01-14 13:25 try using R-C bandpass in clock 2011-01-14 13:27 okay, that should be easy 2011-01-14 13:28 -C-|R|-C-|R|---|C|-R-|C|-R- 2011-01-14 13:28 huh ? not 2011-01-14 13:28 1st 4 are the highpass, followed by a lowpass 2011-01-14 13:28 --C--+--R--+-- 2011-01-14 13:29 .Â Â RÂ Â Â Â C 2011-01-14 13:29 ? 2011-01-14 13:29 (hmm, i wonder how this looks) 2011-01-14 13:29 http://www.electronics-tutorials.ws/filter/filter_4.html 2011-01-14 13:29 you have about twice the number of components i can comfortably fit :-( 2011-01-14 13:30 well, but then my circuit is twice as good :-D 2011-01-14 13:30 you may get away with a single RC for high and low 2011-01-14 13:31 (single RC for high/low) how's that ? 2011-01-14 13:31 or one RC for lowpass and 2 RC for highpass 2011-01-14 13:32 depends on how bad your clock is contaminated. And nota bene R-C filters won't fix jitter 2011-01-14 13:35 i think i'll try RC+RC. that still looks reasonable. for anything more complex, i'd basically poke around blindly, because any measurement would be likely to derail things 2011-01-14 13:35 how wide should i make the pass band ? 2011-01-14 13:35 also consider finding a bead that has a resonance peak @16MHz 2011-01-14 13:35 (wide) -0 2011-01-14 13:35 so far, beads have done absolutely nothing. (don't have a large choice of them, though) 2011-01-14 13:36 I have a complete component book, with maybe 200 different breed 2011-01-14 13:36 I'l have a look 2011-01-14 13:37 placing such a bead instead of R2 in your link above could increase filter quality noticably 2011-01-14 13:39 the beads i have are (digi-key number) 587-1866-1-nd, 587-2419-1-nd, and 587-1926-1-nd. plus a few 490-4012-1-nd arrays. 2011-01-14 13:39 you'd calculate with R2 = bead's off-resonance impedance 2011-01-14 13:40 hmm, that's usually something very close to 0 Ohm 2011-01-14 13:41 i understand what you're saying. not sure about that :) 2011-01-14 13:41 e.g., consider these: http://www.yuden.co.jp/ut/product/pdf/mlci08_e.pdf 2011-01-14 13:43 let's say BKP1005HS330. that one has 33 Ohm at 100 MHz. close to 0 Ohm at DC. or would i use the impedance at the clock frequency, 16 MHz ? that would be something like 15-20 Ohm. 2011-01-14 13:44 or maybe the BK1608HM601, with 600 Ohm at 100 MHz ? 2011-01-14 13:45 nah, beads usually have something like 33R or 100R, for AC. DC isn't relevant here 2011-01-14 13:46 but isn't that the resonance impedance ? 2011-01-14 13:46 alas all my FBM-11 beads here have a peak next 100MHz, or maybe 30MHz (one of them) 2011-01-14 13:47 none has a usable characteristics for 16MHz 2011-01-14 13:47 yup, here too 2011-01-14 13:53 FBM-11-321611-202: R @16MHz: ~1000R,Â Â @10:500R,Â Â @40:2700, @2:~10..50 (too small on this diagram) 2011-01-14 13:54 not from digi-key, i suppose ? 2011-01-14 13:54 at least i can't see it there 2011-01-14 13:55 R(DC): 1.2ohm;Â Â I(DC):100mA; size: 3.2*1.6*1.1mm 2011-01-14 13:55 Richco 2011-01-14 13:56 (3.2*1.6 mm) does this come with a SUV for transportation ? ;-) 2011-01-14 13:57 wow. 1206. and i was cursing those bulky 0603 beads :) 2011-01-14 13:58 naw, that's waaaay to big. 2011-01-14 13:58 s/to/too/ 2011-01-14 13:58 you won't get magic ferrite 2011-01-14 13:59 anyway, I guess I can't help any better 2011-01-14 14:00 as mentioned above, that's way beyond my competence 2011-01-14 14:00 good luck 2011-01-14 14:00 small inductors would be an option 2011-01-14 14:01 beads don't seem to fit well in general. i guess for such special needs, one would use a T or Pi filter instead of a bead. 2011-01-14 14:03 well perhaps just RC then. if that doesn't work, then with a bloody crystal. 2011-01-14 14:04 for RC, how large a pass band would you recommend ? or should i define it in terms of the attenuation of the clock frequency ? 2011-01-14 14:04 and pass band symmetric around the clock frequency ? 2011-01-14 14:12 I'd guess you want especially low freq cutoff. I don't expect freq>16MHz to introduce much noise relevant for clock generation. So calculate R-C for -6dB @ 16MHz (i.e f0 >clock). For the lowpass you probably are free to choose anything that does a semidecent job at attenuating harmonics 2011-01-14 14:13 I'd even consider a 2nd order highpass raher than a bandpass, with a given number of components 2011-01-14 14:14 but as mentioned before, all that won't help to kill jitter 2011-01-14 14:14 jitter in turn will cause FM, which easily creates the sidebands you see 2011-01-14 14:14 a high-pass may at least kill jitter caused by noise riding on the signal and pushing the edges 2011-01-14 14:14 yes 2011-01-14 14:14 that's the idea 2011-01-14 14:15 hmm, the noise is only about +/- 1 MHz 2011-01-14 14:15 so expect your noise on clock being in same freq range 2011-01-14 14:16 of course, that's after the clock input circuit and after the PLL. cause and effect are a bit distant here ... 2011-01-14 14:16 though maybe not, it's a complex circuit, including a PLL to upscale clock I guess 2011-01-14 14:16 yep :) 2011-01-14 14:17 does the PLL have any external components? like R, C? 2011-01-14 14:17 nope. it's all inside 2011-01-14 14:18 hmm nope 2011-01-14 14:18 the worse effect you'll see of any jitter 2011-01-14 14:20 actually .. if i go for -7.5 dB at 16 MHz, then i don't need a divider. one component saved :) 2011-01-14 14:20 the PLL probably is *very* fast, and each single 16MHz clock edge with a tiny bit of jitter causing massive FM noise on TX 2011-01-14 14:24 [commit] kyak: Tile: small fix. http://qi-hw.com/p/openwrt-packages/e5d0837 2011-01-14 14:24 let's just hope the ben doesn't jitter much :) 2011-01-14 14:28 ooh, btw: werner, you are feeding the clock via a long cable, from mainboard to atusd? That probably won't pan out anyway, without using symmetric feed and a balun at both ends 2011-01-14 14:29 about 7 cm in total 2011-01-14 14:30 (cpu to transceiver) 2011-01-14 14:30 my board sits in the uSD slot 2011-01-14 14:32 might work 2011-01-14 14:32 keep trace pseudo-symmetric to a parallel ground though 2011-01-14 14:34 and of course impedance tuned 2011-01-14 14:35 you don't want any ringing and/or standing waves on that line 2011-01-14 14:35 plus you want it properly shielded, also against crosstalk from traces next to it 2011-01-14 14:36 i have no control over what's going on in the ben ... 2011-01-14 14:36 SD clock is between the power lines, so this helps 2011-01-14 14:37 that might be a problem then, disqualifying this clock signal for use as a TX source, with high requirements for S/N and jitter free 2011-01-14 14:38 SD clock? Hell! that's by design not meant to be jitter free 2011-01-14 14:38 honestly, go for XTAL 2011-01-14 14:40 I thought you're using a master clock source, like CPU XTAL or sth 2011-01-14 14:40 not something generated in a SD card interface, and never meant to be jitter free or anything you'd want to see for TX clock source 2011-01-14 14:41 why would the SD clock have more jitter than the master clock it's derived from ? 2011-01-14 14:41 it's all just dividers 2011-01-14 14:41 because there are lots of dividers and switches and latches and whatnot in between 2011-01-14 14:42 all of those not properly shielded in-chip, as jitter free never been a design parameter for that clock 2011-01-14 14:43 hmm, i would naively assume that a divider by itself is fairly jitter-free 2011-01-14 14:44 and a devider isn'r per se jitter free 2011-01-14 14:44 shielding, i don't know, of course 2011-01-14 14:44 but then, considering that beading the clock didn't make any difference, it seems that it's not overly dirty 2011-01-14 14:45 nope, a divider may count up to a limit, then reset. The reset for sure may have diferent properties with signal delay than the normal increment. Just as one simple example of imanent jitter 2011-01-14 14:45 then there's also crosstalk jitter 2011-01-14 14:46 (reset) sure, but the output is always the reset, never the increment 2011-01-14 14:46 (well, never the increment-sans-reset) 2011-01-14 14:46 the output might even be latched by a completely unrelated clock 2011-01-14 14:47 honestly, you can't assume any quality regarding jitter whatsoever, from a storage or similar clock 2011-01-14 14:47 big maybe. usually the whole clock tree is derived from the same source. of course, there may be fun effects if different paths merge 2011-01-14 14:47 it's digital domain, while your TX clock basically is analog domain 2011-01-14 14:48 well .. i'll try my luck with the rc bandpass. if it works, praise to ingenic's diligent engineers :) if not, all hail our crystalline savior ;-) 2011-01-14 14:50 requirements for digital are just-in-time, and correct sequence aka propagation delay. Requirements for analog are amplitude and signal waveform stability, jitter free, and low freq error 2011-01-14 14:58 well, it's a digital output. so if the signal driving it isn't too crazy, it would have a clean waveform. this is a simple chip, so i wouldn't expect a lot of fancy detours that could introduce problems. but let's see ... 2011-01-14 15:11 simulation says, in a configuration like this: http://www.play-hookey.com/ac_theory/band_pass_filters.html 2011-01-14 15:12 100 pF, 68 Ohm, 150 Ohm, 100 pF, from left to right, should yield reasonable results 2011-01-14 15:13 do these values sound appropriate or should i aim for entirely different values ? (still need to do the analytical verification) 2011-01-14 15:14 that's even starting at such simple things like chip internal VDD, which is the digital VDD for this output, incl all the internally generated noise on it. For a proper external clock generator you'd include a choke and a buffer C to decouple the clock VDD from rest of circuit, as on a clock with high requirements on jitter etc you don't tolerated the typical digital amplitude variations (high: 0.7..1.0 * VDD; low: GND - 0.7V .. 0.3 * 2011-01-14 15:14 VDD)) 2011-01-14 15:16 (R, C) hmm, sounds reasonable. Just make sure your input and output impedances are way below (in) rsp above (out) those values 2011-01-14 15:17 out, no problem. for input, i actually tried to be reasonably close to what i expect the driver's impedance. but lemme check if anything is specified ... 2011-01-14 15:17 or, alternatively, include any known impedance into your calculation 2011-01-14 15:18 ssd clock is specified at 4 mA .. that's ... 2011-01-14 15:21 Ioh = 10.2 mA (typ), Voh = 2.4 V, i.e., a 900 mV drop, so Ron = 88.2 Ohm. hmm. 2011-01-14 15:22 you said way below. let's make my resistor a bit larger then ... 2011-01-14 15:23 there, 47 pF and 220 Ohm 2011-01-14 15:29 trick question: what do you have to do when the "continuous wave" looks like this: http://downloads.qi-hardware.com/people/werner/tmp/ghost.png ? 2011-01-14 15:29 answer: power-cycle the USRP ;-) 2011-01-14 15:29 ugh 2011-01-14 15:31 it likes to play these little games. not sure what's happening. at least now i know that i should always reset it before starting any measurements. before figuring out that part, i had been chasing some 20 dB mystery differences for days ... 2011-01-14 15:31 calculate your highpass so it will work with and without a R of (here) 88.2R in series to C1 2011-01-14 15:36 yeah, looks good. peak shifts a little bit to lower frequencies, but not excessively. maybe 1-2 MHz. 2011-01-14 15:36 naw, more. 3-4 MHz. hard to read those logarithmic diagrams :) 2011-01-14 15:36 anyway, it's flat up there. so not a big change. and the rest stays pretty much the same. 2011-01-14 15:39 time for 12648430 2011-01-14 15:40 nice :) 2011-01-14 15:41 here, it's time for laundry-in and trash-out. then a bit of food, rework, and then we'll see whether it's champagne or a bottle of cheap booze :) 2011-01-14 15:42 I'd rather stick with some 48879 2011-01-14 15:47 steve|m: GOOD choice ! ;-) 2011-01-14 15:49 plus 720942 :) 2011-01-14 15:50 48879, hmm yummy 2011-01-14 15:53 I once picked vid/pid 0xB00D 0xC0DE for an usb bootloader ;) 2011-01-14 15:54 but 0xBOO7 0xL04D is nice, too :) 2011-01-14 15:59 Hi all. 2011-01-14 16:03 Has anyone here used Atanua or similar logic simulator software? 2011-01-14 16:12 thanks steve|m for the idea :) 2011-01-14 16:29 wpwrak: do you know of a FOSS way for me to put virtual logic circuits together and test them without learning verilog? 2011-01-14 16:31 qbject: you may want to look through the (free) tools assembled in the fedora electronic lab 2011-01-14 16:31 http://spins.fedoraproject.org/fel/#portfolio 2011-01-14 16:31 toped, magic, alliance, gtkwave, etc. 2011-01-14 16:32 Sweeeet. 2011-01-14 16:32 thanks! 2011-01-14 16:33 yeah, sure. report your experience back here :-) 2011-01-14 16:33 I know pretty much nothing about these tools either, except that they exist. 2011-01-14 16:35 I have some more work to do in KiCad/boom/brdhist/QCad 2011-01-14 16:35 get to IC stuff later... 2011-01-14 16:35 Will do. Might be a while before I become competent to assess, as I'm simultaneously reading up on CPU design and FORTH 2011-01-14 16:37 DocScrutinizer: sigh. last entry of http://downloads.qi-hardware.com/people/werner/wpan/rework-20110113/ 2011-01-14 16:38 qbject: (logic circuits) i don't know much about these. i use qucs fairly extensively to simulate analog circuits, though. it should also be able to do digital. 2011-01-14 16:38 wolfspraul: are you not sleeping any more? 2011-01-14 16:40 I'm in Germany right now, returning to China in a few days though. 2011-01-14 16:40 wpwrak: Also sweet. As long as it can do the usual gates, I'll be good. 2011-01-14 16:40 wolfspraul: Ahh, cool. Good trip? 2011-01-14 16:44 qbject: to Germany? yes sure. Good 27c3 conference, good new contacts in Berlin. 2011-01-14 16:45 DocScrutinizer: either your jitter theory has indeed some merit or the time-space continuum simply refuses to be bent more than 30 dB out of shape around the ben 2011-01-14 16:46 next stop: crystal. the moment of truth. 2011-01-14 16:49 DocScrutinizer: in any case, thanks a lot for your help ! it's also reassuring to know that i didn't miss anything obvious elsewhere. 2011-01-14 16:57 wolfspraul: sounds like an amazing get-together. 2011-01-14 18:25 ola werner 2011-01-14 18:25 wpwrak: what are you trying to do? (bend time?) 2011-01-14 18:25 http://www.flickr.com/photos/41832965@N00/4258134502/ 2011-01-14 18:34 roh: heh ;-) 2011-01-14 18:34 roh: naw, the idea is to get the wpan stuff to work on the ben reusing the ben's clock (and crystal) 2011-01-14 18:34 roh: unfortunately, i found out that there's a fair bit of noise and that noise seems to come from the clock. 2011-01-14 18:37 roh: alas, i can't tell with certainty, also due to equipment limitations. the next iteration should bring more clarity. if a board with a crystal has the same noise pattern, then the noise comes from somewhere else. if not, then we have the culprit. 2011-01-14 18:38 roh: at the same time, the USB-based board looks good. i haven't done any BER testing yet, but the spectrum and rssi patterns all look healthy. 2011-01-14 18:38 nice 2011-01-14 18:38 dont overoptimize ;) 2011-01-14 18:38 roh: BER is probably also the only thing that's easy to check for production testing. 2011-01-14 18:39 meaning.. is using another crystal that bad for you? 2011-01-14 18:39 roh: ah, i don't like crystals. also, the crystal makes the thing bigger. 2011-01-14 18:40 roh: the price difference is negligible for the small quantities we're looking at at the moment. but still, it's nice to make the best of it. 2011-01-14 18:41 hehe.. as long as the part exists its good and easy 2011-01-14 18:41 i failed at buying black screws 2011-01-14 18:41 .s 2011-01-14 18:42 now ive ordered the ST2.2 metal screws in 'blank, zinc' since i can only buy sizes ST2.9 from upwards in black 2011-01-14 18:42 you can always paint them :) 2011-01-14 18:43 a can of epoxy paint goes a long way :) 2011-01-14 18:43 naah. that scratches of and sticks into the small places where you would like to apply the screwdriver ;) 2011-01-14 18:44 the only proper way to color screws seems to be galvanisation 2011-01-14 18:44 (small places) i think the screwdriver will just win :) 2011-01-14 18:44 but yes, the stuff comes off 2011-01-14 18:48 roh: you there? 2011-01-14 18:49 just a heads up, you know I ordered 10 violet cases from you, so that's moving... 2011-01-14 18:49 but I wanted to let you know I am thinking about rc3, and my current plan is maybe 80 units 2011-01-14 18:49 roh: all sleep has been cancelled until easter :) 2011-01-14 18:49 so that would mean another 80 violet cases. but that's later. don't order now, I will ponder some more. 2011-01-14 18:50 no no, just way in advance. just so you've heard that number already. 80. 2011-01-14 18:50 it will be between 50 and 100, I am leaning towards 80 now. 2011-01-14 18:50 and I want cases for all of them, and only offer fully assembled units. 2011-01-14 18:51 for cheap things like screws and such you can already source 100 sets, if that even makes any significant price difference. 2011-01-14 18:51 but for the more expensive stuff, don't source yet, I will only order those 80 in a month or so, if all goes well. 2011-01-14 18:52 let's do the next 10 first :-) 2011-01-14 18:57 wow 80 ! 2011-01-14 19:00 kristianpaul: ah yeah, better don't ask. hardware is painful. 2011-01-14 19:00 once you start, you just have to continue and slowly increase the volume. 2011-01-14 19:00 that applied to then Ben Nanonote as well? 2011-01-14 19:01 I have no choice but to pull forward, make a complete product out of it (case, accessories, certification, box, manual, etc). 2011-01-14 19:01 oh sure, even more so 2011-01-14 19:01 and unless I have 50 or so in stock, it makes no sense to do a launch 2011-01-14 19:01 otherwise the launch press release goes out, some media pick it up, but I cannot deliver 2011-01-14 19:01 even 50 isn't a lot :) 2011-01-14 19:02 of course if nobody picks it up, and I have no marketing power to sell them, I get stuck with all the junk 2011-01-14 19:02 better delete that from the logs ;-) 2011-01-14 19:02 did you made a launh for the ben nanonote? 2011-01-14 19:02 if I sell for 499 USD finding 50 actual paying customers is a nice achievement 2011-01-14 19:02 I'm not deleting any of this from the logs. 2011-01-14 19:02 (just kidding) 2011-01-14 19:03 I'm just saying I cannot make a launch if I have 10 in stock. 2011-01-14 19:03 it all makes no sense, then I am wasting my time, and the chance to say 'new product launched', and cannot deliver. 2011-01-14 19:03 it's all risky 2011-01-14 19:03